1 /*
   2  * Copyright (c) 2003, 2011, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/assembler.hpp"
  27 #include "interpreter/bytecodeHistogram.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "interpreter/interpreterGenerator.hpp"
  30 #include "interpreter/interpreterRuntime.hpp"
  31 #include "interpreter/templateTable.hpp"
  32 #include "oops/arrayOop.hpp"
  33 #include "oops/methodDataOop.hpp"
  34 #include "oops/methodOop.hpp"
  35 #include "oops/oop.inline.hpp"
  36 #include "prims/jvmtiExport.hpp"
  37 #include "prims/jvmtiThreadState.hpp"
  38 #include "runtime/arguments.hpp"
  39 #include "runtime/deoptimization.hpp"
  40 #include "runtime/frame.inline.hpp"
  41 #include "runtime/sharedRuntime.hpp"
  42 #include "runtime/stubRoutines.hpp"
  43 #include "runtime/synchronizer.hpp"
  44 #include "runtime/timer.hpp"
  45 #include "runtime/vframeArray.hpp"
  46 #include "utilities/debug.hpp"
  47 
  48 #define __ _masm->
  49 
  50 #ifndef CC_INTERP
  51 
  52 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
  53 const int bci_offset    = frame::interpreter_frame_bcx_offset    * wordSize;
  54 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
  55 
  56 //-----------------------------------------------------------------------------
  57 
  58 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
  59   address entry = __ pc();
  60 
  61 #ifdef ASSERT
  62   {
  63     Label L;
  64     __ lea(rax, Address(rbp,
  65                         frame::interpreter_frame_monitor_block_top_offset *
  66                         wordSize));
  67     __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
  68                          // grows negative)
  69     __ jcc(Assembler::aboveEqual, L); // check if frame is complete
  70     __ stop ("interpreter frame not set up");
  71     __ bind(L);
  72   }
  73 #endif // ASSERT
  74   // Restore bcp under the assumption that the current frame is still
  75   // interpreted
  76   __ restore_bcp();
  77 
  78   // expression stack must be empty before entering the VM if an
  79   // exception happened
  80   __ empty_expression_stack();
  81   // throw exception
  82   __ call_VM(noreg,
  83              CAST_FROM_FN_PTR(address,
  84                               InterpreterRuntime::throw_StackOverflowError));
  85   return entry;
  86 }
  87 
  88 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
  89         const char* name) {
  90   address entry = __ pc();
  91   // expression stack must be empty before entering the VM if an
  92   // exception happened
  93   __ empty_expression_stack();
  94   // setup parameters
  95   // ??? convention: expect aberrant index in register ebx
  96   __ lea(c_rarg1, ExternalAddress((address)name));
  97   __ call_VM(noreg,
  98              CAST_FROM_FN_PTR(address,
  99                               InterpreterRuntime::
 100                               throw_ArrayIndexOutOfBoundsException),
 101              c_rarg1, rbx);
 102   return entry;
 103 }
 104 
 105 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 106   address entry = __ pc();
 107 
 108   // object is at TOS
 109   __ pop(c_rarg1);
 110 
 111   // expression stack must be empty before entering the VM if an
 112   // exception happened
 113   __ empty_expression_stack();
 114 
 115   __ call_VM(noreg,
 116              CAST_FROM_FN_PTR(address,
 117                               InterpreterRuntime::
 118                               throw_ClassCastException),
 119              c_rarg1);
 120   return entry;
 121 }
 122 
 123 address TemplateInterpreterGenerator::generate_exception_handler_common(
 124         const char* name, const char* message, bool pass_oop) {
 125   assert(!pass_oop || message == NULL, "either oop or message but not both");
 126   address entry = __ pc();
 127   if (pass_oop) {
 128     // object is at TOS
 129     __ pop(c_rarg2);
 130   }
 131   // expression stack must be empty before entering the VM if an
 132   // exception happened
 133   __ empty_expression_stack();
 134   // setup parameters
 135   __ lea(c_rarg1, ExternalAddress((address)name));
 136   if (pass_oop) {
 137     __ call_VM(rax, CAST_FROM_FN_PTR(address,
 138                                      InterpreterRuntime::
 139                                      create_klass_exception),
 140                c_rarg1, c_rarg2);
 141   } else {
 142     // kind of lame ExternalAddress can't take NULL because
 143     // external_word_Relocation will assert.
 144     if (message != NULL) {
 145       __ lea(c_rarg2, ExternalAddress((address)message));
 146     } else {
 147       __ movptr(c_rarg2, NULL_WORD);
 148     }
 149     __ call_VM(rax,
 150                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 151                c_rarg1, c_rarg2);
 152   }
 153   // throw exception
 154   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 155   return entry;
 156 }
 157 
 158 
 159 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
 160   address entry = __ pc();
 161   // NULL last_sp until next java call
 162   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 163   __ dispatch_next(state);
 164   return entry;
 165 }
 166 
 167 
 168 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
 169   address entry = __ pc();
 170 
 171   // Restore stack bottom in case i2c adjusted stack
 172   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 173   // and NULL it as marker that esp is now tos until next java call
 174   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 175 
 176   __ restore_bcp();
 177   __ restore_locals();
 178 
 179   Label L_got_cache, L_giant_index;
 180   if (EnableInvokeDynamic) {
 181     __ cmpb(Address(r13, 0), Bytecodes::_invokedynamic);
 182     __ jcc(Assembler::equal, L_giant_index);
 183   }
 184   __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
 185   __ bind(L_got_cache);
 186   __ movl(rbx, Address(rbx, rcx,
 187                        Address::times_ptr,
 188                        in_bytes(constantPoolCacheOopDesc::base_offset()) +
 189                        3 * wordSize));
 190   __ andl(rbx, 0xFF);
 191   __ lea(rsp, Address(rsp, rbx, Address::times_8));
 192   __ dispatch_next(state, step);
 193 
 194   // out of the main line of code...
 195   if (EnableInvokeDynamic) {
 196     __ bind(L_giant_index);
 197     __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
 198     __ jmp(L_got_cache);
 199   }
 200 
 201   return entry;
 202 }
 203 
 204 
 205 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
 206                                                                int step) {
 207   address entry = __ pc();
 208   // NULL last_sp until next java call
 209   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 210   __ restore_bcp();
 211   __ restore_locals();
 212   // handle exceptions
 213   {
 214     Label L;
 215     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
 216     __ jcc(Assembler::zero, L);
 217     __ call_VM(noreg,
 218                CAST_FROM_FN_PTR(address,
 219                                 InterpreterRuntime::throw_pending_exception));
 220     __ should_not_reach_here();
 221     __ bind(L);
 222   }
 223   __ dispatch_next(state, step);
 224   return entry;
 225 }
 226 
 227 int AbstractInterpreter::BasicType_as_index(BasicType type) {
 228   int i = 0;
 229   switch (type) {
 230     case T_BOOLEAN: i = 0; break;
 231     case T_CHAR   : i = 1; break;
 232     case T_BYTE   : i = 2; break;
 233     case T_SHORT  : i = 3; break;
 234     case T_INT    : i = 4; break;
 235     case T_LONG   : i = 5; break;
 236     case T_VOID   : i = 6; break;
 237     case T_FLOAT  : i = 7; break;
 238     case T_DOUBLE : i = 8; break;
 239     case T_OBJECT : i = 9; break;
 240     case T_ARRAY  : i = 9; break;
 241     default       : ShouldNotReachHere();
 242   }
 243   assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
 244          "index out of bounds");
 245   return i;
 246 }
 247 
 248 
 249 address TemplateInterpreterGenerator::generate_result_handler_for(
 250         BasicType type) {
 251   address entry = __ pc();
 252   switch (type) {
 253   case T_BOOLEAN: __ c2bool(rax);            break;
 254   case T_CHAR   : __ movzwl(rax, rax);       break;
 255   case T_BYTE   : __ sign_extend_byte(rax);  break;
 256   case T_SHORT  : __ sign_extend_short(rax); break;
 257   case T_INT    : /* nothing to do */        break;
 258   case T_LONG   : /* nothing to do */        break;
 259   case T_VOID   : /* nothing to do */        break;
 260   case T_FLOAT  : /* nothing to do */        break;
 261   case T_DOUBLE : /* nothing to do */        break;
 262   case T_OBJECT :
 263     // retrieve result from frame
 264     __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 265     // and verify it
 266     __ verify_oop(rax);
 267     break;
 268   default       : ShouldNotReachHere();
 269   }
 270   __ ret(0);                                   // return from result handler
 271   return entry;
 272 }
 273 
 274 address TemplateInterpreterGenerator::generate_safept_entry_for(
 275         TosState state,
 276         address runtime_entry) {
 277   address entry = __ pc();
 278   __ push(state);
 279   __ call_VM(noreg, runtime_entry);
 280   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 281   return entry;
 282 }
 283 
 284 
 285 
 286 // Helpers for commoning out cases in the various type of method entries.
 287 //
 288 
 289 
 290 // increment invocation count & check for overflow
 291 //
 292 // Note: checking for negative value instead of overflow
 293 //       so we have a 'sticky' overflow test
 294 //
 295 // rbx: method
 296 // ecx: invocation counter
 297 //
 298 void InterpreterGenerator::generate_counter_incr(
 299         Label* overflow,
 300         Label* profile_method,
 301         Label* profile_method_continue) {
 302   const Address invocation_counter(rbx, in_bytes(methodOopDesc::invocation_counter_offset()) +
 303                                         in_bytes(InvocationCounter::counter_offset()));
 304   // Note: In tiered we increment either counters in methodOop or in MDO depending if we're profiling or not.
 305   if (TieredCompilation) {
 306     int increment = InvocationCounter::count_increment;
 307     int mask = ((1 << Tier0InvokeNotifyFreqLog)  - 1) << InvocationCounter::count_shift;
 308     Label no_mdo, done;
 309     if (ProfileInterpreter) {
 310       // Are we profiling?
 311       __ movptr(rax, Address(rbx, methodOopDesc::method_data_offset()));
 312       __ testptr(rax, rax);
 313       __ jccb(Assembler::zero, no_mdo);
 314       // Increment counter in the MDO
 315       const Address mdo_invocation_counter(rax, in_bytes(methodDataOopDesc::invocation_counter_offset()) +
 316                                                 in_bytes(InvocationCounter::counter_offset()));
 317       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
 318       __ jmpb(done);
 319     }
 320     __ bind(no_mdo);
 321     // Increment counter in methodOop (we don't need to load it, it's in ecx).
 322     __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow);
 323     __ bind(done);
 324   } else {
 325     const Address backedge_counter(rbx,
 326                                    methodOopDesc::backedge_counter_offset() +
 327                                    InvocationCounter::counter_offset());
 328 
 329     if (ProfileInterpreter) { // %%% Merge this into methodDataOop
 330       __ incrementl(Address(rbx,
 331                             methodOopDesc::interpreter_invocation_counter_offset()));
 332     }
 333     // Update standard invocation counters
 334     __ movl(rax, backedge_counter);   // load backedge counter
 335 
 336     __ incrementl(rcx, InvocationCounter::count_increment);
 337     __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
 338 
 339     __ movl(invocation_counter, rcx); // save invocation count
 340     __ addl(rcx, rax);                // add both counters
 341 
 342     // profile_method is non-null only for interpreted method so
 343     // profile_method != NULL == !native_call
 344 
 345     if (ProfileInterpreter && profile_method != NULL) {
 346       // Test to see if we should create a method data oop
 347       __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
 348       __ jcc(Assembler::less, *profile_method_continue);
 349 
 350       // if no method data exists, go to profile_method
 351       __ test_method_data_pointer(rax, *profile_method);
 352     }
 353 
 354     __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
 355     __ jcc(Assembler::aboveEqual, *overflow);
 356   }
 357 }
 358 
 359 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
 360 
 361   // Asm interpreter on entry
 362   // r14 - locals
 363   // r13 - bcp
 364   // rbx - method
 365   // edx - cpool --- DOES NOT APPEAR TO BE TRUE
 366   // rbp - interpreter frame
 367 
 368   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 369   // Everything as it was on entry
 370   // rdx is not restored. Doesn't appear to really be set.
 371 
 372   const Address size_of_parameters(rbx,
 373                                    methodOopDesc::size_of_parameters_offset());
 374 
 375   // InterpreterRuntime::frequency_counter_overflow takes two
 376   // arguments, the first (thread) is passed by call_VM, the second
 377   // indicates if the counter overflow occurs at a backwards branch
 378   // (NULL bcp).  We pass zero for it.  The call returns the address
 379   // of the verified entry point for the method or NULL if the
 380   // compilation did not complete (either went background or bailed
 381   // out).
 382   __ movl(c_rarg1, 0);
 383   __ call_VM(noreg,
 384              CAST_FROM_FN_PTR(address,
 385                               InterpreterRuntime::frequency_counter_overflow),
 386              c_rarg1);
 387 
 388   __ movptr(rbx, Address(rbp, method_offset));   // restore methodOop
 389   // Preserve invariant that r13/r14 contain bcp/locals of sender frame
 390   // and jump to the interpreted entry.
 391   __ jmp(*do_continue, relocInfo::none);
 392 }
 393 
 394 // See if we've got enough room on the stack for locals plus overhead.
 395 // The expression stack grows down incrementally, so the normal guard
 396 // page mechanism will work for that.
 397 //
 398 // NOTE: Since the additional locals are also always pushed (wasn't
 399 // obvious in generate_method_entry) so the guard should work for them
 400 // too.
 401 //
 402 // Args:
 403 //      rdx: number of additional locals this frame needs (what we must check)
 404 //      rbx: methodOop
 405 //
 406 // Kills:
 407 //      rax
 408 void InterpreterGenerator::generate_stack_overflow_check(void) {
 409 
 410   // monitor entry size: see picture of stack set
 411   // (generate_method_entry) and frame_amd64.hpp
 412   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 413 
 414   // total overhead size: entry_size + (saved rbp through expr stack
 415   // bottom).  be sure to change this if you add/subtract anything
 416   // to/from the overhead area
 417   const int overhead_size =
 418     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 419 
 420   const int page_size = os::vm_page_size();
 421 
 422   Label after_frame_check;
 423 
 424   // see if the frame is greater than one page in size. If so,
 425   // then we need to verify there is enough stack space remaining
 426   // for the additional locals.
 427   __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
 428   __ jcc(Assembler::belowEqual, after_frame_check);
 429 
 430   // compute rsp as if this were going to be the last frame on
 431   // the stack before the red zone
 432 
 433   const Address stack_base(r15_thread, Thread::stack_base_offset());
 434   const Address stack_size(r15_thread, Thread::stack_size_offset());
 435 
 436   // locals + overhead, in bytes
 437   __ mov(rax, rdx);
 438   __ shlptr(rax, Interpreter::logStackElementSize);  // 2 slots per parameter.
 439   __ addptr(rax, overhead_size);
 440 
 441 #ifdef ASSERT
 442   Label stack_base_okay, stack_size_okay;
 443   // verify that thread stack base is non-zero
 444   __ cmpptr(stack_base, (int32_t)NULL_WORD);
 445   __ jcc(Assembler::notEqual, stack_base_okay);
 446   __ stop("stack base is zero");
 447   __ bind(stack_base_okay);
 448   // verify that thread stack size is non-zero
 449   __ cmpptr(stack_size, 0);
 450   __ jcc(Assembler::notEqual, stack_size_okay);
 451   __ stop("stack size is zero");
 452   __ bind(stack_size_okay);
 453 #endif
 454 
 455   // Add stack base to locals and subtract stack size
 456   __ addptr(rax, stack_base);
 457   __ subptr(rax, stack_size);
 458 
 459   // Use the maximum number of pages we might bang.
 460   const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
 461                                                                               (StackRedPages+StackYellowPages);
 462 
 463   // add in the red and yellow zone sizes
 464   __ addptr(rax, max_pages * page_size);
 465 
 466   // check against the current stack bottom
 467   __ cmpptr(rsp, rax);
 468   __ jcc(Assembler::above, after_frame_check);
 469 
 470   __ pop(rax); // get return address
 471   __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry()));
 472 
 473   // all done with frame size check
 474   __ bind(after_frame_check);
 475 }
 476 
 477 // Allocate monitor and lock method (asm interpreter)
 478 //
 479 // Args:
 480 //      rbx: methodOop
 481 //      r14: locals
 482 //
 483 // Kills:
 484 //      rax
 485 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 486 //      rscratch1, rscratch2 (scratch regs)
 487 void InterpreterGenerator::lock_method(void) {
 488   // synchronize method
 489   const Address access_flags(rbx, methodOopDesc::access_flags_offset());
 490   const Address monitor_block_top(
 491         rbp,
 492         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 493   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 494 
 495 #ifdef ASSERT
 496   {
 497     Label L;
 498     __ movl(rax, access_flags);
 499     __ testl(rax, JVM_ACC_SYNCHRONIZED);
 500     __ jcc(Assembler::notZero, L);
 501     __ stop("method doesn't need synchronization");
 502     __ bind(L);
 503   }
 504 #endif // ASSERT
 505 
 506   // get synchronization object
 507   {
 508     const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() +
 509                               Klass::java_mirror_offset_in_bytes();
 510     Label done;
 511     __ movl(rax, access_flags);
 512     __ testl(rax, JVM_ACC_STATIC);
 513     // get receiver (assume this is frequent case)
 514     __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0)));
 515     __ jcc(Assembler::zero, done);
 516     __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
 517     __ movptr(rax, Address(rax,
 518                            constantPoolOopDesc::pool_holder_offset_in_bytes()));
 519     __ movptr(rax, Address(rax, mirror_offset));
 520 
 521 #ifdef ASSERT
 522     {
 523       Label L;
 524       __ testptr(rax, rax);
 525       __ jcc(Assembler::notZero, L);
 526       __ stop("synchronization object is NULL");
 527       __ bind(L);
 528     }
 529 #endif // ASSERT
 530 
 531     __ bind(done);
 532   }
 533 
 534   // add space for monitor & lock
 535   __ subptr(rsp, entry_size); // add space for a monitor entry
 536   __ movptr(monitor_block_top, rsp);  // set new monitor block top
 537   // store object
 538   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
 539   __ movptr(c_rarg1, rsp); // object address
 540   __ lock_object(c_rarg1);
 541 }
 542 
 543 // Generate a fixed interpreter frame. This is identical setup for
 544 // interpreted methods and for native methods hence the shared code.
 545 //
 546 // Args:
 547 //      rax: return address
 548 //      rbx: methodOop
 549 //      r14: pointer to locals
 550 //      r13: sender sp
 551 //      rdx: cp cache
 552 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 553   // initialize fixed part of activation frame
 554   __ push(rax);        // save return address
 555   __ enter();          // save old & set new rbp
 556   __ push(r13);        // set sender sp
 557   __ push((int)NULL_WORD); // leave last_sp as null
 558   __ movptr(r13, Address(rbx, methodOopDesc::const_offset()));      // get constMethodOop
 559   __ lea(r13, Address(r13, constMethodOopDesc::codes_offset())); // get codebase
 560   __ push(rbx);        // save methodOop
 561   if (ProfileInterpreter) {
 562     Label method_data_continue;
 563     __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
 564     __ testptr(rdx, rdx);
 565     __ jcc(Assembler::zero, method_data_continue);
 566     __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset()));
 567     __ bind(method_data_continue);
 568     __ push(rdx);      // set the mdp (method data pointer)
 569   } else {
 570     __ push(0);
 571   }
 572 
 573   __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset()));
 574   __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
 575   __ push(rdx); // set constant pool cache
 576   __ push(r14); // set locals pointer
 577   if (native_call) {
 578     __ push(0); // no bcp
 579   } else {
 580     __ push(r13); // set bcp
 581   }
 582   __ push(0); // reserve word for pointer to expression stack bottom
 583   __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
 584 }
 585 
 586 // End of helpers
 587 
 588 // Various method entries
 589 //------------------------------------------------------------------------------------------------------------------------
 590 //
 591 //
 592 
 593 // Call an accessor method (assuming it is resolved, otherwise drop
 594 // into vanilla (slow path) entry
 595 address InterpreterGenerator::generate_accessor_entry(void) {
 596   // rbx: methodOop
 597 
 598   // r13: senderSP must preserver for slow path, set SP to it on fast path
 599 
 600   address entry_point = __ pc();
 601   Label xreturn_path;
 602 
 603   // do fastpath for resolved accessor methods
 604   if (UseFastAccessorMethods) {
 605     // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites
 606     //       thereof; parameter size = 1
 607     // Note: We can only use this code if the getfield has been resolved
 608     //       and if we don't have a null-pointer exception => check for
 609     //       these conditions first and use slow path if necessary.
 610     Label slow_path;
 611     // If we need a safepoint check, generate full interpreter entry.
 612     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 613              SafepointSynchronize::_not_synchronized);
 614 
 615     __ jcc(Assembler::notEqual, slow_path);
 616     // rbx: method
 617     __ movptr(rax, Address(rsp, wordSize));
 618 
 619     // check if local 0 != NULL and read field
 620     __ testptr(rax, rax);
 621     __ jcc(Assembler::zero, slow_path);
 622 
 623     __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset()));
 624     // read first instruction word and extract bytecode @ 1 and index @ 2
 625     __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
 626     __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
 627     // Shift codes right to get the index on the right.
 628     // The bytecode fetched looks like <index><0xb4><0x2a>
 629     __ shrl(rdx, 2 * BitsPerByte);
 630     __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
 631     __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
 632 
 633     // rax: local 0
 634     // rbx: method
 635     // rdx: constant pool cache index
 636     // rdi: constant pool cache
 637 
 638     // check if getfield has been resolved and read constant pool cache entry
 639     // check the validity of the cache entry by testing whether _indices field
 640     // contains Bytecode::_getfield in b1 byte.
 641     assert(in_words(ConstantPoolCacheEntry::size()) == 4,
 642            "adjust shift below");
 643     __ movl(rcx,
 644             Address(rdi,
 645                     rdx,
 646                     Address::times_8,
 647                     constantPoolCacheOopDesc::base_offset() +
 648                     ConstantPoolCacheEntry::indices_offset()));
 649     __ shrl(rcx, 2 * BitsPerByte);
 650     __ andl(rcx, 0xFF);
 651     __ cmpl(rcx, Bytecodes::_getfield);
 652     __ jcc(Assembler::notEqual, slow_path);
 653 
 654     // Note: constant pool entry is not valid before bytecode is resolved
 655     __ movptr(rcx,
 656               Address(rdi,
 657                       rdx,
 658                       Address::times_8,
 659                       constantPoolCacheOopDesc::base_offset() +
 660                       ConstantPoolCacheEntry::f2_offset()));
 661     // edx: flags
 662     __ movl(rdx,
 663             Address(rdi,
 664                     rdx,
 665                     Address::times_8,
 666                     constantPoolCacheOopDesc::base_offset() +
 667                     ConstantPoolCacheEntry::flags_offset()));
 668 
 669     Label notObj, notInt, notByte, notShort;
 670     const Address field_address(rax, rcx, Address::times_1);
 671 
 672     // Need to differentiate between igetfield, agetfield, bgetfield etc.
 673     // because they are different sizes.
 674     // Use the type from the constant pool cache
 675     __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
 676     // Make sure we don't need to mask edx for tosBits after the above shift
 677     ConstantPoolCacheEntry::verify_tosBits();
 678 
 679     __ cmpl(rdx, atos);
 680     __ jcc(Assembler::notEqual, notObj);
 681     // atos
 682     __ load_heap_oop(rax, field_address);
 683     __ jmp(xreturn_path);
 684 
 685     __ bind(notObj);
 686     __ cmpl(rdx, itos);
 687     __ jcc(Assembler::notEqual, notInt);
 688     // itos
 689     __ movl(rax, field_address);
 690     __ jmp(xreturn_path);
 691 
 692     __ bind(notInt);
 693     __ cmpl(rdx, btos);
 694     __ jcc(Assembler::notEqual, notByte);
 695     // btos
 696     __ load_signed_byte(rax, field_address);
 697     __ jmp(xreturn_path);
 698 
 699     __ bind(notByte);
 700     __ cmpl(rdx, stos);
 701     __ jcc(Assembler::notEqual, notShort);
 702     // stos
 703     __ load_signed_short(rax, field_address);
 704     __ jmp(xreturn_path);
 705 
 706     __ bind(notShort);
 707 #ifdef ASSERT
 708     Label okay;
 709     __ cmpl(rdx, ctos);
 710     __ jcc(Assembler::equal, okay);
 711     __ stop("what type is this?");
 712     __ bind(okay);
 713 #endif
 714     // ctos
 715     __ load_unsigned_short(rax, field_address);
 716 
 717     __ bind(xreturn_path);
 718 
 719     // _ireturn/_areturn
 720     __ pop(rdi);
 721     __ mov(rsp, r13);
 722     __ jmp(rdi);
 723     __ ret(0);
 724 
 725     // generate a vanilla interpreter entry as the slow path
 726     __ bind(slow_path);
 727     (void) generate_normal_entry(false);
 728   } else {
 729     (void) generate_normal_entry(false);
 730   }
 731 
 732   return entry_point;
 733 }
 734 
 735 // Method entry for java.lang.ref.Reference.get.
 736 address InterpreterGenerator::generate_Reference_get_entry(void) {
 737 #ifndef SERIALGC
 738   // Code: _aload_0, _getfield, _areturn
 739   // parameter size = 1
 740   //
 741   // The code that gets generated by this routine is split into 2 parts:
 742   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 743   //    2. The slow path - which is an expansion of the regular method entry.
 744   //
 745   // Notes:-
 746   // * In the G1 code we do not check whether we need to block for
 747   //   a safepoint. If G1 is enabled then we must execute the specialized
 748   //   code for Reference.get (except when the Reference object is null)
 749   //   so that we can log the value in the referent field with an SATB
 750   //   update buffer.
 751   //   If the code for the getfield template is modified so that the
 752   //   G1 pre-barrier code is executed when the current method is
 753   //   Reference.get() then going through the normal method entry
 754   //   will be fine.
 755   // * The G1 code can, however, check the receiver object (the instance
 756   //   of java.lang.Reference) and jump to the slow path if null. If the
 757   //   Reference object is null then we obviously cannot fetch the referent
 758   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 759   //   regular method entry code to generate the NPE.
 760   //
 761   // This code is based on generate_accessor_enty.
 762   //
 763   // rbx: methodOop
 764 
 765   // r13: senderSP must preserve for slow path, set SP to it on fast path
 766 
 767   address entry = __ pc();
 768 
 769   const int referent_offset = java_lang_ref_Reference::referent_offset;
 770   guarantee(referent_offset > 0, "referent offset not initialized");
 771 
 772   if (UseG1GC) {
 773     Label slow_path;
 774     // rbx: method
 775 
 776     // Check if local 0 != NULL
 777     // If the receiver is null then it is OK to jump to the slow path.
 778     __ movptr(rax, Address(rsp, wordSize));
 779 
 780     __ testptr(rax, rax);
 781     __ jcc(Assembler::zero, slow_path);
 782 
 783     // rax: local 0
 784     // rbx: method (but can be used as scratch now)
 785     // rdx: scratch
 786     // rdi: scratch
 787 
 788     // Generate the G1 pre-barrier code to log the value of
 789     // the referent field in an SATB buffer.
 790 
 791     // Load the value of the referent field.
 792     const Address field_address(rax, referent_offset);
 793     __ load_heap_oop(rax, field_address);
 794 
 795     // Generate the G1 pre-barrier code to log the value of
 796     // the referent field in an SATB buffer.
 797     __ g1_write_barrier_pre(noreg /* obj */,
 798                             rax /* pre_val */,
 799                             r15_thread /* thread */,
 800                             rbx /* tmp */,
 801                             true /* tosca_live */,
 802                             true /* expand_call */);
 803 
 804     // _areturn
 805     __ pop(rdi);                // get return address
 806     __ mov(rsp, r13);           // set sp to sender sp
 807     __ jmp(rdi);
 808     __ ret(0);
 809 
 810     // generate a vanilla interpreter entry as the slow path
 811     __ bind(slow_path);
 812     (void) generate_normal_entry(false);
 813 
 814     return entry;
 815   }
 816 #endif // SERIALGC
 817 
 818   // If G1 is not enabled then attempt to go through the accessor entry point
 819   // Reference.get is an accessor
 820   return generate_accessor_entry();
 821 }
 822 
 823 
 824 // Interpreter stub for calling a native method. (asm interpreter)
 825 // This sets up a somewhat different looking stack for calling the
 826 // native method than the typical interpreter frame setup.
 827 address InterpreterGenerator::generate_native_entry(bool synchronized) {
 828   // determine code generation flags
 829   bool inc_counter  = UseCompiler || CountCompiledCalls;
 830 
 831   // rbx: methodOop
 832   // r13: sender sp
 833 
 834   address entry_point = __ pc();
 835 
 836   const Address size_of_parameters(rbx, methodOopDesc::
 837                                         size_of_parameters_offset());
 838   const Address invocation_counter(rbx, methodOopDesc::
 839                                         invocation_counter_offset() +
 840                                         InvocationCounter::counter_offset());
 841   const Address access_flags      (rbx, methodOopDesc::access_flags_offset());
 842 
 843   // get parameter size (always needed)
 844   __ load_unsigned_short(rcx, size_of_parameters);
 845 
 846   // native calls don't need the stack size check since they have no
 847   // expression stack and the arguments are already on the stack and
 848   // we only add a handful of words to the stack
 849 
 850   // rbx: methodOop
 851   // rcx: size of parameters
 852   // r13: sender sp
 853   __ pop(rax);                                       // get return address
 854 
 855   // for natives the size of locals is zero
 856 
 857   // compute beginning of parameters (r14)
 858   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
 859 
 860   // add 2 zero-initialized slots for native calls
 861   // initialize result_handler slot
 862   __ push((int) NULL_WORD);
 863   // slot for oop temp
 864   // (static native method holder mirror/jni oop result)
 865   __ push((int) NULL_WORD);
 866 
 867   if (inc_counter) {
 868     __ movl(rcx, invocation_counter);  // (pre-)fetch invocation count
 869   }
 870 
 871   // initialize fixed part of activation frame
 872   generate_fixed_frame(true);
 873 
 874   // make sure method is native & not abstract
 875 #ifdef ASSERT
 876   __ movl(rax, access_flags);
 877   {
 878     Label L;
 879     __ testl(rax, JVM_ACC_NATIVE);
 880     __ jcc(Assembler::notZero, L);
 881     __ stop("tried to execute non-native method as native");
 882     __ bind(L);
 883   }
 884   {
 885     Label L;
 886     __ testl(rax, JVM_ACC_ABSTRACT);
 887     __ jcc(Assembler::zero, L);
 888     __ stop("tried to execute abstract method in interpreter");
 889     __ bind(L);
 890   }
 891 #endif
 892 
 893   // Since at this point in the method invocation the exception handler
 894   // would try to exit the monitor of synchronized methods which hasn't
 895   // been entered yet, we set the thread local variable
 896   // _do_not_unlock_if_synchronized to true. The remove_activation will
 897   // check this flag.
 898 
 899   const Address do_not_unlock_if_synchronized(r15_thread,
 900         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 901   __ movbool(do_not_unlock_if_synchronized, true);
 902 
 903   // increment invocation count & check for overflow
 904   Label invocation_counter_overflow;
 905   if (inc_counter) {
 906     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
 907   }
 908 
 909   Label continue_after_compile;
 910   __ bind(continue_after_compile);
 911 
 912   bang_stack_shadow_pages(true);
 913 
 914   // reset the _do_not_unlock_if_synchronized flag
 915   __ movbool(do_not_unlock_if_synchronized, false);
 916 
 917   // check for synchronized methods
 918   // Must happen AFTER invocation_counter check and stack overflow check,
 919   // so method is not locked if overflows.
 920   if (synchronized) {
 921     lock_method();
 922   } else {
 923     // no synchronization necessary
 924 #ifdef ASSERT
 925     {
 926       Label L;
 927       __ movl(rax, access_flags);
 928       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 929       __ jcc(Assembler::zero, L);
 930       __ stop("method needs synchronization");
 931       __ bind(L);
 932     }
 933 #endif
 934   }
 935 
 936   // start execution
 937 #ifdef ASSERT
 938   {
 939     Label L;
 940     const Address monitor_block_top(rbp,
 941                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 942     __ movptr(rax, monitor_block_top);
 943     __ cmpptr(rax, rsp);
 944     __ jcc(Assembler::equal, L);
 945     __ stop("broken stack frame setup in interpreter");
 946     __ bind(L);
 947   }
 948 #endif
 949 
 950   // jvmti support
 951   __ notify_method_entry();
 952 
 953   // work registers
 954   const Register method = rbx;
 955   const Register t      = r11;
 956 
 957   // allocate space for parameters
 958   __ get_method(method);
 959   __ verify_oop(method);
 960   __ load_unsigned_short(t,
 961                          Address(method,
 962                                  methodOopDesc::size_of_parameters_offset()));
 963   __ shll(t, Interpreter::logStackElementSize);
 964 
 965   __ subptr(rsp, t);
 966   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
 967   __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
 968 
 969   // get signature handler
 970   {
 971     Label L;
 972     __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
 973     __ testptr(t, t);
 974     __ jcc(Assembler::notZero, L);
 975     __ call_VM(noreg,
 976                CAST_FROM_FN_PTR(address,
 977                                 InterpreterRuntime::prepare_native_call),
 978                method);
 979     __ get_method(method);
 980     __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
 981     __ bind(L);
 982   }
 983 
 984   // call signature handler
 985   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14,
 986          "adjust this code");
 987   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
 988          "adjust this code");
 989   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
 990           "adjust this code");
 991 
 992   // The generated handlers do not touch RBX (the method oop).
 993   // However, large signatures cannot be cached and are generated
 994   // each time here.  The slow-path generator can do a GC on return,
 995   // so we must reload it after the call.
 996   __ call(t);
 997   __ get_method(method);        // slow path can do a GC, reload RBX
 998 
 999 
1000   // result handler is in rax
1001   // set result handler
1002   __ movptr(Address(rbp,
1003                     (frame::interpreter_frame_result_handler_offset) * wordSize),
1004             rax);
1005 
1006   // pass mirror handle if static call
1007   {
1008     Label L;
1009     const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() +
1010                               Klass::java_mirror_offset_in_bytes();
1011     __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1012     __ testl(t, JVM_ACC_STATIC);
1013     __ jcc(Assembler::zero, L);
1014     // get mirror
1015     __ movptr(t, Address(method, methodOopDesc::constants_offset()));
1016     __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
1017     __ movptr(t, Address(t, mirror_offset));
1018     // copy mirror into activation frame
1019     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1020             t);
1021     // pass handle to mirror
1022     __ lea(c_rarg1,
1023            Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1024     __ bind(L);
1025   }
1026 
1027   // get native function entry point
1028   {
1029     Label L;
1030     __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
1031     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1032     __ movptr(rscratch2, unsatisfied.addr());
1033     __ cmpptr(rax, rscratch2);
1034     __ jcc(Assembler::notEqual, L);
1035     __ call_VM(noreg,
1036                CAST_FROM_FN_PTR(address,
1037                                 InterpreterRuntime::prepare_native_call),
1038                method);
1039     __ get_method(method);
1040     __ verify_oop(method);
1041     __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
1042     __ bind(L);
1043   }
1044 
1045   // pass JNIEnv
1046   __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1047 
1048   // It is enough that the pc() points into the right code
1049   // segment. It does not have to be the correct return pc.
1050   __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1051 
1052   // change thread state
1053 #ifdef ASSERT
1054   {
1055     Label L;
1056     __ movl(t, Address(r15_thread, JavaThread::thread_state_offset()));
1057     __ cmpl(t, _thread_in_Java);
1058     __ jcc(Assembler::equal, L);
1059     __ stop("Wrong thread state in native stub");
1060     __ bind(L);
1061   }
1062 #endif
1063 
1064   // Change state to native
1065 
1066   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1067           _thread_in_native);
1068 
1069   // Call the native method.
1070   __ call(rax);
1071   // result potentially in rax or xmm0
1072 
1073   // Depending on runtime options, either restore the MXCSR
1074   // register after returning from the JNI Call or verify that
1075   // it wasn't changed during -Xcheck:jni.
1076   if (RestoreMXCSROnJNICalls) {
1077     __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
1078   }
1079   else if (CheckJNICalls) {
1080     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry())));
1081   }
1082 
1083   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1084   // in order to extract the result of a method call. If the order of these
1085   // pushes change or anything else is added to the stack then the code in
1086   // interpreter_frame_result must also change.
1087 
1088   __ push(dtos);
1089   __ push(ltos);
1090 
1091   // change thread state
1092   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1093           _thread_in_native_trans);
1094 
1095   if (os::is_MP()) {
1096     if (UseMembar) {
1097       // Force this write out before the read below
1098       __ membar(Assembler::Membar_mask_bits(
1099            Assembler::LoadLoad | Assembler::LoadStore |
1100            Assembler::StoreLoad | Assembler::StoreStore));
1101     } else {
1102       // Write serialization page so VM thread can do a pseudo remote membar.
1103       // We use the current thread pointer to calculate a thread specific
1104       // offset to write to within the page. This minimizes bus traffic
1105       // due to cache line collision.
1106       __ serialize_memory(r15_thread, rscratch2);
1107     }
1108   }
1109 
1110   // check for safepoint operation in progress and/or pending suspend requests
1111   {
1112     Label Continue;
1113     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1114              SafepointSynchronize::_not_synchronized);
1115 
1116     Label L;
1117     __ jcc(Assembler::notEqual, L);
1118     __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1119     __ jcc(Assembler::equal, Continue);
1120     __ bind(L);
1121 
1122     // Don't use call_VM as it will see a possible pending exception
1123     // and forward it and never return here preventing us from
1124     // clearing _last_native_pc down below.  Also can't use
1125     // call_VM_leaf either as it will check to see if r13 & r14 are
1126     // preserved and correspond to the bcp/locals pointers. So we do a
1127     // runtime call by hand.
1128     //
1129     __ mov(c_rarg0, r15_thread);
1130     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1131     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1132     __ andptr(rsp, -16); // align stack as required by ABI
1133     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1134     __ mov(rsp, r12); // restore sp
1135     __ reinit_heapbase();
1136     __ bind(Continue);
1137   }
1138 
1139   // change thread state
1140   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1141 
1142   // reset_last_Java_frame
1143   __ reset_last_Java_frame(true, true);
1144 
1145   // reset handle block
1146   __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset()));
1147   __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1148 
1149   // If result is an oop unbox and store it in frame where gc will see it
1150   // and result handler will pick it up
1151 
1152   {
1153     Label no_oop, store_result;
1154     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1155     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1156     __ jcc(Assembler::notEqual, no_oop);
1157     // retrieve result
1158     __ pop(ltos);
1159     __ testptr(rax, rax);
1160     __ jcc(Assembler::zero, store_result);
1161     __ movptr(rax, Address(rax, 0));
1162     __ bind(store_result);
1163     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1164     // keep stack depth as expected by pushing oop which will eventually be discarde
1165     __ push(ltos);
1166     __ bind(no_oop);
1167   }
1168 
1169 
1170   {
1171     Label no_reguard;
1172     __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()),
1173             JavaThread::stack_guard_yellow_disabled);
1174     __ jcc(Assembler::notEqual, no_reguard);
1175 
1176     __ pusha(); // XXX only save smashed registers
1177     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1178     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1179     __ andptr(rsp, -16); // align stack as required by ABI
1180     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1181     __ mov(rsp, r12); // restore sp
1182     __ popa(); // XXX only restore smashed registers
1183     __ reinit_heapbase();
1184 
1185     __ bind(no_reguard);
1186   }
1187 
1188 
1189   // The method register is junk from after the thread_in_native transition
1190   // until here.  Also can't call_VM until the bcp has been
1191   // restored.  Need bcp for throwing exception below so get it now.
1192   __ get_method(method);
1193   __ verify_oop(method);
1194 
1195   // restore r13 to have legal interpreter frame, i.e., bci == 0 <=>
1196   // r13 == code_base()
1197   __ movptr(r13, Address(method, methodOopDesc::const_offset()));   // get constMethodOop
1198   __ lea(r13, Address(r13, constMethodOopDesc::codes_offset()));    // get codebase
1199   // handle exceptions (exception handling will handle unlocking!)
1200   {
1201     Label L;
1202     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1203     __ jcc(Assembler::zero, L);
1204     // Note: At some point we may want to unify this with the code
1205     // used in call_VM_base(); i.e., we should use the
1206     // StubRoutines::forward_exception code. For now this doesn't work
1207     // here because the rsp is not correctly set at this point.
1208     __ MacroAssembler::call_VM(noreg,
1209                                CAST_FROM_FN_PTR(address,
1210                                InterpreterRuntime::throw_pending_exception));
1211     __ should_not_reach_here();
1212     __ bind(L);
1213   }
1214 
1215   // do unlocking if necessary
1216   {
1217     Label L;
1218     __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1219     __ testl(t, JVM_ACC_SYNCHRONIZED);
1220     __ jcc(Assembler::zero, L);
1221     // the code below should be shared with interpreter macro
1222     // assembler implementation
1223     {
1224       Label unlock;
1225       // BasicObjectLock will be first in list, since this is a
1226       // synchronized method. However, need to check that the object
1227       // has not been unlocked by an explicit monitorexit bytecode.
1228       const Address monitor(rbp,
1229                             (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1230                                        wordSize - sizeof(BasicObjectLock)));
1231 
1232       // monitor expect in c_rarg1 for slow unlock path
1233       __ lea(c_rarg1, monitor); // address of first monitor
1234 
1235       __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1236       __ testptr(t, t);
1237       __ jcc(Assembler::notZero, unlock);
1238 
1239       // Entry already unlocked, need to throw exception
1240       __ MacroAssembler::call_VM(noreg,
1241                                  CAST_FROM_FN_PTR(address,
1242                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1243       __ should_not_reach_here();
1244 
1245       __ bind(unlock);
1246       __ unlock_object(c_rarg1);
1247     }
1248     __ bind(L);
1249   }
1250 
1251   // jvmti support
1252   // Note: This must happen _after_ handling/throwing any exceptions since
1253   //       the exception handler code notifies the runtime of method exits
1254   //       too. If this happens before, method entry/exit notifications are
1255   //       not properly paired (was bug - gri 11/22/99).
1256   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1257 
1258   // restore potential result in edx:eax, call result handler to
1259   // restore potential result in ST0 & handle result
1260 
1261   __ pop(ltos);
1262   __ pop(dtos);
1263 
1264   __ movptr(t, Address(rbp,
1265                        (frame::interpreter_frame_result_handler_offset) * wordSize));
1266   __ call(t);
1267 
1268   // remove activation
1269   __ movptr(t, Address(rbp,
1270                        frame::interpreter_frame_sender_sp_offset *
1271                        wordSize)); // get sender sp
1272   __ leave();                                // remove frame anchor
1273   __ pop(rdi);                               // get return address
1274   __ mov(rsp, t);                            // set sp to sender sp
1275   __ jmp(rdi);
1276 
1277   if (inc_counter) {
1278     // Handle overflow of counter and compile method
1279     __ bind(invocation_counter_overflow);
1280     generate_counter_overflow(&continue_after_compile);
1281   }
1282 
1283   return entry_point;
1284 }
1285 
1286 //
1287 // Generic interpreted method entry to (asm) interpreter
1288 //
1289 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1290   // determine code generation flags
1291   bool inc_counter  = UseCompiler || CountCompiledCalls;
1292 
1293   // ebx: methodOop
1294   // r13: sender sp
1295   address entry_point = __ pc();
1296 
1297   const Address size_of_parameters(rbx,
1298                                    methodOopDesc::size_of_parameters_offset());
1299   const Address size_of_locals(rbx, methodOopDesc::size_of_locals_offset());
1300   const Address invocation_counter(rbx,
1301                                    methodOopDesc::invocation_counter_offset() +
1302                                    InvocationCounter::counter_offset());
1303   const Address access_flags(rbx, methodOopDesc::access_flags_offset());
1304 
1305   // get parameter size (always needed)
1306   __ load_unsigned_short(rcx, size_of_parameters);
1307 
1308   // rbx: methodOop
1309   // rcx: size of parameters
1310   // r13: sender_sp (could differ from sp+wordSize if we were called via c2i )
1311 
1312   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1313   __ subl(rdx, rcx); // rdx = no. of additional locals
1314 
1315   // YYY
1316 //   __ incrementl(rdx);
1317 //   __ andl(rdx, -2);
1318 
1319   // see if we've got enough room on the stack for locals plus overhead.
1320   generate_stack_overflow_check();
1321 
1322   // get return address
1323   __ pop(rax);
1324 
1325   // compute beginning of parameters (r14)
1326   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1327 
1328   // rdx - # of additional locals
1329   // allocate space for locals
1330   // explicitly initialize locals
1331   {
1332     Label exit, loop;
1333     __ testl(rdx, rdx);
1334     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1335     __ bind(loop);
1336     __ push((int) NULL_WORD); // initialize local variables
1337     __ decrementl(rdx); // until everything initialized
1338     __ jcc(Assembler::greater, loop);
1339     __ bind(exit);
1340   }
1341 
1342   // (pre-)fetch invocation count
1343   if (inc_counter) {
1344     __ movl(rcx, invocation_counter);
1345   }
1346   // initialize fixed part of activation frame
1347   generate_fixed_frame(false);
1348 
1349   // make sure method is not native & not abstract
1350 #ifdef ASSERT
1351   __ movl(rax, access_flags);
1352   {
1353     Label L;
1354     __ testl(rax, JVM_ACC_NATIVE);
1355     __ jcc(Assembler::zero, L);
1356     __ stop("tried to execute native method as non-native");
1357     __ bind(L);
1358   }
1359   {
1360     Label L;
1361     __ testl(rax, JVM_ACC_ABSTRACT);
1362     __ jcc(Assembler::zero, L);
1363     __ stop("tried to execute abstract method in interpreter");
1364     __ bind(L);
1365   }
1366 #endif
1367 
1368   // Since at this point in the method invocation the exception
1369   // handler would try to exit the monitor of synchronized methods
1370   // which hasn't been entered yet, we set the thread local variable
1371   // _do_not_unlock_if_synchronized to true. The remove_activation
1372   // will check this flag.
1373 
1374   const Address do_not_unlock_if_synchronized(r15_thread,
1375         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1376   __ movbool(do_not_unlock_if_synchronized, true);
1377 
1378   // increment invocation count & check for overflow
1379   Label invocation_counter_overflow;
1380   Label profile_method;
1381   Label profile_method_continue;
1382   if (inc_counter) {
1383     generate_counter_incr(&invocation_counter_overflow,
1384                           &profile_method,
1385                           &profile_method_continue);
1386     if (ProfileInterpreter) {
1387       __ bind(profile_method_continue);
1388     }
1389   }
1390 
1391   Label continue_after_compile;
1392   __ bind(continue_after_compile);
1393 
1394   // check for synchronized interpreted methods
1395   bang_stack_shadow_pages(false);
1396 
1397   // reset the _do_not_unlock_if_synchronized flag
1398   __ movbool(do_not_unlock_if_synchronized, false);
1399 
1400   // check for synchronized methods
1401   // Must happen AFTER invocation_counter check and stack overflow check,
1402   // so method is not locked if overflows.
1403   if (synchronized) {
1404     // Allocate monitor and lock method
1405     lock_method();
1406   } else {
1407     // no synchronization necessary
1408 #ifdef ASSERT
1409     {
1410       Label L;
1411       __ movl(rax, access_flags);
1412       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1413       __ jcc(Assembler::zero, L);
1414       __ stop("method needs synchronization");
1415       __ bind(L);
1416     }
1417 #endif
1418   }
1419 
1420   // start execution
1421 #ifdef ASSERT
1422   {
1423     Label L;
1424      const Address monitor_block_top (rbp,
1425                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1426     __ movptr(rax, monitor_block_top);
1427     __ cmpptr(rax, rsp);
1428     __ jcc(Assembler::equal, L);
1429     __ stop("broken stack frame setup in interpreter");
1430     __ bind(L);
1431   }
1432 #endif
1433 
1434   // jvmti support
1435   __ notify_method_entry();
1436 
1437   __ dispatch_next(vtos);
1438 
1439   // invocation counter overflow
1440   if (inc_counter) {
1441     if (ProfileInterpreter) {
1442       // We have decided to profile this method in the interpreter
1443       __ bind(profile_method);
1444       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1445       __ set_method_data_pointer_for_bcp();
1446       __ get_method(rbx);
1447       __ jmp(profile_method_continue);
1448     }
1449     // Handle overflow of counter and compile method
1450     __ bind(invocation_counter_overflow);
1451     generate_counter_overflow(&continue_after_compile);
1452   }
1453 
1454   return entry_point;
1455 }
1456 
1457 // Entry points
1458 //
1459 // Here we generate the various kind of entries into the interpreter.
1460 // The two main entry type are generic bytecode methods and native
1461 // call method.  These both come in synchronized and non-synchronized
1462 // versions but the frame layout they create is very similar. The
1463 // other method entry types are really just special purpose entries
1464 // that are really entry and interpretation all in one. These are for
1465 // trivial methods like accessor, empty, or special math methods.
1466 //
1467 // When control flow reaches any of the entry types for the interpreter
1468 // the following holds ->
1469 //
1470 // Arguments:
1471 //
1472 // rbx: methodOop
1473 //
1474 // Stack layout immediately at entry
1475 //
1476 // [ return address     ] <--- rsp
1477 // [ parameter n        ]
1478 //   ...
1479 // [ parameter 1        ]
1480 // [ expression stack   ] (caller's java expression stack)
1481 
1482 // Assuming that we don't go to one of the trivial specialized entries
1483 // the stack will look like below when we are ready to execute the
1484 // first bytecode (or call the native routine). The register usage
1485 // will be as the template based interpreter expects (see
1486 // interpreter_amd64.hpp).
1487 //
1488 // local variables follow incoming parameters immediately; i.e.
1489 // the return address is moved to the end of the locals).
1490 //
1491 // [ monitor entry      ] <--- rsp
1492 //   ...
1493 // [ monitor entry      ]
1494 // [ expr. stack bottom ]
1495 // [ saved r13          ]
1496 // [ current r14        ]
1497 // [ methodOop          ]
1498 // [ saved ebp          ] <--- rbp
1499 // [ return address     ]
1500 // [ local variable m   ]
1501 //   ...
1502 // [ local variable 1   ]
1503 // [ parameter n        ]
1504 //   ...
1505 // [ parameter 1        ] <--- r14
1506 
1507 address AbstractInterpreterGenerator::generate_method_entry(
1508                                         AbstractInterpreter::MethodKind kind) {
1509   // determine code generation flags
1510   bool synchronized = false;
1511   address entry_point = NULL;
1512 
1513   switch (kind) {
1514   case Interpreter::zerolocals             :                                                                             break;
1515   case Interpreter::zerolocals_synchronized: synchronized = true;                                                        break;
1516   case Interpreter::native                 : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); break;
1517   case Interpreter::native_synchronized    : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(true);  break;
1518   case Interpreter::empty                  : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry();       break;
1519   case Interpreter::accessor               : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry();    break;
1520   case Interpreter::abstract               : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry();    break;
1521   case Interpreter::method_handle          : entry_point = ((InterpreterGenerator*) this)->generate_method_handle_entry();break;
1522 
1523   case Interpreter::java_lang_math_sin     : // fall thru
1524   case Interpreter::java_lang_math_cos     : // fall thru
1525   case Interpreter::java_lang_math_tan     : // fall thru
1526   case Interpreter::java_lang_math_abs     : // fall thru
1527   case Interpreter::java_lang_math_log     : // fall thru
1528   case Interpreter::java_lang_math_log10   : // fall thru
1529   case Interpreter::java_lang_math_sqrt    : entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind);    break;
1530   case Interpreter::java_lang_ref_reference_get
1531                                            : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1532   default                                  : ShouldNotReachHere();                                                       break;
1533   }
1534 
1535   if (entry_point) {
1536     return entry_point;
1537   }
1538 
1539   return ((InterpreterGenerator*) this)->
1540                                 generate_normal_entry(synchronized);
1541 }
1542 
1543 // These should never be compiled since the interpreter will prefer
1544 // the compiled version to the intrinsic version.
1545 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1546   switch (method_kind(m)) {
1547     case Interpreter::java_lang_math_sin     : // fall thru
1548     case Interpreter::java_lang_math_cos     : // fall thru
1549     case Interpreter::java_lang_math_tan     : // fall thru
1550     case Interpreter::java_lang_math_abs     : // fall thru
1551     case Interpreter::java_lang_math_log     : // fall thru
1552     case Interpreter::java_lang_math_log10   : // fall thru
1553     case Interpreter::java_lang_math_sqrt    :
1554       return false;
1555     default:
1556       return true;
1557   }
1558 }
1559 
1560 // How much stack a method activation needs in words.
1561 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
1562   const int entry_size = frame::interpreter_frame_monitor_size();
1563 
1564   // total overhead size: entry_size + (saved rbp thru expr stack
1565   // bottom).  be sure to change this if you add/subtract anything
1566   // to/from the overhead area
1567   const int overhead_size =
1568     -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1569 
1570   const int stub_code = frame::entry_frame_after_call_words;
1571   const int extra_stack = methodOopDesc::extra_stack_entries();
1572   const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1573                            Interpreter::stackElementWords;
1574   return (overhead_size + method_stack + stub_code);
1575 }
1576 
1577 int AbstractInterpreter::layout_activation(methodOop method,
1578                                            int tempcount,
1579                                            int popframe_extra_args,
1580                                            int moncount,
1581                                            int caller_actual_parameters,
1582                                            int callee_param_count,
1583                                            int callee_locals,
1584                                            frame* caller,
1585                                            frame* interpreter_frame,
1586                                            bool is_top_frame) {
1587   // Note: This calculation must exactly parallel the frame setup
1588   // in AbstractInterpreterGenerator::generate_method_entry.
1589   // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1590   // The frame interpreter_frame, if not NULL, is guaranteed to be the
1591   // right size, as determined by a previous call to this method.
1592   // It is also guaranteed to be walkable even though it is in a skeletal state
1593 
1594   // fixed size of an interpreter frame:
1595   int max_locals = method->max_locals() * Interpreter::stackElementWords;
1596   int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1597                      Interpreter::stackElementWords;
1598 
1599   int overhead = frame::sender_sp_offset -
1600                  frame::interpreter_frame_initial_sp_offset;
1601   // Our locals were accounted for by the caller (or last_frame_adjust
1602   // on the transistion) Since the callee parameters already account
1603   // for the callee's params we only need to account for the extra
1604   // locals.
1605   int size = overhead +
1606          (callee_locals - callee_param_count)*Interpreter::stackElementWords +
1607          moncount * frame::interpreter_frame_monitor_size() +
1608          tempcount* Interpreter::stackElementWords + popframe_extra_args;
1609   if (interpreter_frame != NULL) {
1610 #ifdef ASSERT
1611     if (!EnableInvokeDynamic)
1612       // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1613       // Probably, since deoptimization doesn't work yet.
1614       assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1615     assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1616 #endif
1617 
1618     interpreter_frame->interpreter_frame_set_method(method);
1619     // NOTE the difference in using sender_sp and
1620     // interpreter_frame_sender_sp interpreter_frame_sender_sp is
1621     // the original sp of the caller (the unextended_sp) and
1622     // sender_sp is fp+16 XXX
1623     intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1624 
1625 #ifdef ASSERT
1626     if (caller->is_interpreted_frame()) {
1627       assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1628     }
1629 #endif
1630 
1631     interpreter_frame->interpreter_frame_set_locals(locals);
1632     BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1633     BasicObjectLock* monbot = montop - moncount;
1634     interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1635 
1636     // Set last_sp
1637     intptr_t*  esp = (intptr_t*) monbot -
1638                      tempcount*Interpreter::stackElementWords -
1639                      popframe_extra_args;
1640     interpreter_frame->interpreter_frame_set_last_sp(esp);
1641 
1642     // All frames but the initial (oldest) interpreter frame we fill in have
1643     // a value for sender_sp that allows walking the stack but isn't
1644     // truly correct. Correct the value here.
1645     if (extra_locals != 0 &&
1646         interpreter_frame->sender_sp() ==
1647         interpreter_frame->interpreter_frame_sender_sp()) {
1648       interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() +
1649                                                          extra_locals);
1650     }
1651     *interpreter_frame->interpreter_frame_cache_addr() =
1652       method->constants()->cache();
1653   }
1654   return size;
1655 }
1656 
1657 //-----------------------------------------------------------------------------
1658 // Exceptions
1659 
1660 void TemplateInterpreterGenerator::generate_throw_exception() {
1661   // Entry point in previous activation (i.e., if the caller was
1662   // interpreted)
1663   Interpreter::_rethrow_exception_entry = __ pc();
1664   // Restore sp to interpreter_frame_last_sp even though we are going
1665   // to empty the expression stack for the exception processing.
1666   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1667   // rax: exception
1668   // rdx: return address/pc that threw exception
1669   __ restore_bcp();    // r13 points to call/send
1670   __ restore_locals();
1671   __ reinit_heapbase();  // restore r12 as heapbase.
1672   // Entry point for exceptions thrown within interpreter code
1673   Interpreter::_throw_exception_entry = __ pc();
1674   // expression stack is undefined here
1675   // rax: exception
1676   // r13: exception bcp
1677   __ verify_oop(rax);
1678   __ mov(c_rarg1, rax);
1679 
1680   // expression stack must be empty before entering the VM in case of
1681   // an exception
1682   __ empty_expression_stack();
1683   // find exception handler address and preserve exception oop
1684   __ call_VM(rdx,
1685              CAST_FROM_FN_PTR(address,
1686                           InterpreterRuntime::exception_handler_for_exception),
1687              c_rarg1);
1688   // rax: exception handler entry point
1689   // rdx: preserved exception oop
1690   // r13: bcp for exception handler
1691   __ push_ptr(rdx); // push exception which is now the only value on the stack
1692   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1693 
1694   // If the exception is not handled in the current frame the frame is
1695   // removed and the exception is rethrown (i.e. exception
1696   // continuation is _rethrow_exception).
1697   //
1698   // Note: At this point the bci is still the bxi for the instruction
1699   // which caused the exception and the expression stack is
1700   // empty. Thus, for any VM calls at this point, GC will find a legal
1701   // oop map (with empty expression stack).
1702 
1703   // In current activation
1704   // tos: exception
1705   // esi: exception bcp
1706 
1707   //
1708   // JVMTI PopFrame support
1709   //
1710 
1711   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1712   __ empty_expression_stack();
1713   // Set the popframe_processing bit in pending_popframe_condition
1714   // indicating that we are currently handling popframe, so that
1715   // call_VMs that may happen later do not trigger new popframe
1716   // handling cycles.
1717   __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset()));
1718   __ orl(rdx, JavaThread::popframe_processing_bit);
1719   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx);
1720 
1721   {
1722     // Check to see whether we are returning to a deoptimized frame.
1723     // (The PopFrame call ensures that the caller of the popped frame is
1724     // either interpreted or compiled and deoptimizes it if compiled.)
1725     // In this case, we can't call dispatch_next() after the frame is
1726     // popped, but instead must save the incoming arguments and restore
1727     // them after deoptimization has occurred.
1728     //
1729     // Note that we don't compare the return PC against the
1730     // deoptimization blob's unpack entry because of the presence of
1731     // adapter frames in C2.
1732     Label caller_not_deoptimized;
1733     __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize));
1734     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1735                                InterpreterRuntime::interpreter_contains), c_rarg1);
1736     __ testl(rax, rax);
1737     __ jcc(Assembler::notZero, caller_not_deoptimized);
1738 
1739     // Compute size of arguments for saving when returning to
1740     // deoptimized caller
1741     __ get_method(rax);
1742     __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::
1743                                                 size_of_parameters_offset())));
1744     __ shll(rax, Interpreter::logStackElementSize);
1745     __ restore_locals(); // XXX do we need this?
1746     __ subptr(r14, rax);
1747     __ addptr(r14, wordSize);
1748     // Save these arguments
1749     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1750                                            Deoptimization::
1751                                            popframe_preserve_args),
1752                           r15_thread, rax, r14);
1753 
1754     __ remove_activation(vtos, rdx,
1755                          /* throw_monitor_exception */ false,
1756                          /* install_monitor_exception */ false,
1757                          /* notify_jvmdi */ false);
1758 
1759     // Inform deoptimization that it is responsible for restoring
1760     // these arguments
1761     __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1762             JavaThread::popframe_force_deopt_reexecution_bit);
1763 
1764     // Continue in deoptimization handler
1765     __ jmp(rdx);
1766 
1767     __ bind(caller_not_deoptimized);
1768   }
1769 
1770   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1771                        /* throw_monitor_exception */ false,
1772                        /* install_monitor_exception */ false,
1773                        /* notify_jvmdi */ false);
1774 
1775   // Finish with popframe handling
1776   // A previous I2C followed by a deoptimization might have moved the
1777   // outgoing arguments further up the stack. PopFrame expects the
1778   // mutations to those outgoing arguments to be preserved and other
1779   // constraints basically require this frame to look exactly as
1780   // though it had previously invoked an interpreted activation with
1781   // no space between the top of the expression stack (current
1782   // last_sp) and the top of stack. Rather than force deopt to
1783   // maintain this kind of invariant all the time we call a small
1784   // fixup routine to move the mutated arguments onto the top of our
1785   // expression stack if necessary.
1786   __ mov(c_rarg1, rsp);
1787   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1788   // PC must point into interpreter here
1789   __ set_last_Java_frame(noreg, rbp, __ pc());
1790   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1791   __ reset_last_Java_frame(true, true);
1792   // Restore the last_sp and null it out
1793   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1794   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1795 
1796   __ restore_bcp();  // XXX do we need this?
1797   __ restore_locals(); // XXX do we need this?
1798   // The method data pointer was incremented already during
1799   // call profiling. We have to restore the mdp for the current bcp.
1800   if (ProfileInterpreter) {
1801     __ set_method_data_pointer_for_bcp();
1802   }
1803 
1804   // Clear the popframe condition flag
1805   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1806           JavaThread::popframe_inactive);
1807 
1808   __ dispatch_next(vtos);
1809   // end of PopFrame support
1810 
1811   Interpreter::_remove_activation_entry = __ pc();
1812 
1813   // preserve exception over this code sequence
1814   __ pop_ptr(rax);
1815   __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax);
1816   // remove the activation (without doing throws on illegalMonitorExceptions)
1817   __ remove_activation(vtos, rdx, false, true, false);
1818   // restore exception
1819   __ movptr(rax, Address(r15_thread, JavaThread::vm_result_offset()));
1820   __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), (int32_t)NULL_WORD);
1821   __ verify_oop(rax);
1822 
1823   // In between activations - previous activation type unknown yet
1824   // compute continuation point - the continuation point expects the
1825   // following registers set up:
1826   //
1827   // rax: exception
1828   // rdx: return address/pc that threw exception
1829   // rsp: expression stack of caller
1830   // rbp: ebp of caller
1831   __ push(rax);                                  // save exception
1832   __ push(rdx);                                  // save return address
1833   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1834                           SharedRuntime::exception_handler_for_return_address),
1835                         r15_thread, rdx);
1836   __ mov(rbx, rax);                              // save exception handler
1837   __ pop(rdx);                                   // restore return address
1838   __ pop(rax);                                   // restore exception
1839   // Note that an "issuing PC" is actually the next PC after the call
1840   __ jmp(rbx);                                   // jump to exception
1841                                                  // handler of caller
1842 }
1843 
1844 
1845 //
1846 // JVMTI ForceEarlyReturn support
1847 //
1848 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1849   address entry = __ pc();
1850 
1851   __ restore_bcp();
1852   __ restore_locals();
1853   __ empty_expression_stack();
1854   __ load_earlyret_value(state);
1855 
1856   __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
1857   Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset());
1858 
1859   // Clear the earlyret state
1860   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1861 
1862   __ remove_activation(state, rsi,
1863                        false, /* throw_monitor_exception */
1864                        false, /* install_monitor_exception */
1865                        true); /* notify_jvmdi */
1866   __ jmp(rsi);
1867 
1868   return entry;
1869 } // end of ForceEarlyReturn support
1870 
1871 
1872 //-----------------------------------------------------------------------------
1873 // Helper for vtos entry point generation
1874 
1875 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1876                                                          address& bep,
1877                                                          address& cep,
1878                                                          address& sep,
1879                                                          address& aep,
1880                                                          address& iep,
1881                                                          address& lep,
1882                                                          address& fep,
1883                                                          address& dep,
1884                                                          address& vep) {
1885   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1886   Label L;
1887   aep = __ pc();  __ push_ptr();  __ jmp(L);
1888   fep = __ pc();  __ push_f();    __ jmp(L);
1889   dep = __ pc();  __ push_d();    __ jmp(L);
1890   lep = __ pc();  __ push_l();    __ jmp(L);
1891   bep = cep = sep =
1892   iep = __ pc();  __ push_i();
1893   vep = __ pc();
1894   __ bind(L);
1895   generate_and_dispatch(t);
1896 }
1897 
1898 
1899 //-----------------------------------------------------------------------------
1900 // Generation of individual instructions
1901 
1902 // helpers for generate_and_dispatch
1903 
1904 
1905 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1906   : TemplateInterpreterGenerator(code) {
1907    generate_all(); // down here so it can be "virtual"
1908 }
1909 
1910 //-----------------------------------------------------------------------------
1911 
1912 // Non-product code
1913 #ifndef PRODUCT
1914 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1915   address entry = __ pc();
1916 
1917   __ push(state);
1918   __ push(c_rarg0);
1919   __ push(c_rarg1);
1920   __ push(c_rarg2);
1921   __ push(c_rarg3);
1922   __ mov(c_rarg2, rax);  // Pass itos
1923 #ifdef _WIN64
1924   __ movflt(xmm3, xmm0); // Pass ftos
1925 #endif
1926   __ call_VM(noreg,
1927              CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
1928              c_rarg1, c_rarg2, c_rarg3);
1929   __ pop(c_rarg3);
1930   __ pop(c_rarg2);
1931   __ pop(c_rarg1);
1932   __ pop(c_rarg0);
1933   __ pop(state);
1934   __ ret(0);                                   // return from result handler
1935 
1936   return entry;
1937 }
1938 
1939 void TemplateInterpreterGenerator::count_bytecode() {
1940   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1941 }
1942 
1943 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1944   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1945 }
1946 
1947 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1948   __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1949   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1950   __ orl(rbx,
1951          ((int) t->bytecode()) <<
1952          BytecodePairHistogram::log2_number_of_codes);
1953   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1954   __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1955   __ incrementl(Address(rscratch1, rbx, Address::times_4));
1956 }
1957 
1958 
1959 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1960   // Call a little run-time stub to avoid blow-up for each bytecode.
1961   // The run-time runtime saves the right registers, depending on
1962   // the tosca in-state for the given template.
1963 
1964   assert(Interpreter::trace_code(t->tos_in()) != NULL,
1965          "entry must have been generated");
1966   __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1967   __ andptr(rsp, -16); // align stack as required by ABI
1968   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1969   __ mov(rsp, r12); // restore sp
1970   __ reinit_heapbase();
1971 }
1972 
1973 
1974 void TemplateInterpreterGenerator::stop_interpreter_at() {
1975   Label L;
1976   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1977            StopInterpreterAt);
1978   __ jcc(Assembler::notEqual, L);
1979   __ int3();
1980   __ bind(L);
1981 }
1982 #endif // !PRODUCT
1983 #endif // ! CC_INTERP